February 20th 2017; Sala Stringa and North building, FBK, Povo, Trento


[9:00-9:30] Introduction [Francesco Profumo, Paolo Tonella] – Sala Stringa

[9:30-10:30] Keynote: My unconventional journey in the AI world, by FBK PhD alumnus Alfio Gliozzo, IBM Research – Sala Stringa

[10:30-11:00] Coffee break – Open space, North building

[11:00-12:30] Poster session (3rd-4th year FBK students) – Open space, North building

1.Biniam Fisseha Demissie, Security Testing of Android Apps for Detecting Permission Re-delegation Vulnerabilities

2.Martina De Sanctis, Design for Adaptation of Distributed Service-based Systems

3.Ivan Donadello, Semantic Image Interpretation

4.Amin Farajian, Domain Adaptation in Neural Machine Translation

5.Anna Feltracco, From lexical opposition to discourse contrast relations: a study with computational method for the Italian language

6.Andrea Gaiardo, Development of Innovative Materials and Substrates for Gas Sensing Technology

7.Simone Ghio, Multifunctional Bio-inspired Surfaces

8.Didem Gundogdu, Emergency Event Detection and Geographic Propagation Prediction Using Mobile Phone Data

9.Aravind Harikumar, An approach to conifer species classification based on crown structure modeling in high density airborne LiDAR data

10.Gunel Jahangirova, Oracle Problem in Software Testing

11.Elena Kalinina, Real time adaptation of the stimulation protocols for neuroscientific studies

12.Reza Khoshkangini, Automatic Generation and Recommendation of Personalized Challenges in Gamification

13.Antonio Marsico, An Intent-based Negotiation Protocol for Application-Aware Network Services

14.Eleonora Mencarini, Haptic Feedback for Learning to Climb

15.Stefano Menini, Agreement and Disagreement: Comparison of Points of View in the Political Domain

16.Yaroslav Nechaev, Social Annotation and User Profiling

17.Federico Pederzolli, Demonstration of a Hybrid SDN/GMPLS Control Plane for Optical Virtual Private Networks with Restoration Capabilities

18.Andrea Pedrielli, Gas adsorption in graphene-based nanostructures

19.Giulio Petrucci, Ontology Learning in the Deep

20.Mirco Ravanelli, Deep Learning for Distant Speech Recognition

21.Giada Sciarretta, AuthZ/N Protocols for Native Apps: Security and Privacy

22.Francesco Segatta, Modelling Photoinduced Events and Non-Linear Spectroscopy in Complex Multi- Chromophoric Systems

23.Yady Tatiana Solano, Advanced Methods for the Analysis of Very High Resolution Multi-Sensor Optical Images Time-Series

24.Rachele Sprugnoli, “Two days we have passed with the ancients...” From Historical Travel Reports to

Information Extraction

25.Serra Sinem Tekiroglu, Exploring Sensorial Features for Metaphor Identification

26.Georgina Tryfou, Time-frequency reassignment for acoustic signal processing

Lunch [12:30-14:00]


[14:00-15:30] Keynote: European Research Council - Starting Grants: Best ideas for the best researchers by Marco Ferraro, APRE – Sala Stringa

[15:30-16:00] Coffee break – Sala Stringa

[16:00-17:30] Short presentations (students graduating in Spring 2017) – Sala Stringa

1.Danilo Benozzo, Detecting brain effective connectivity with supervised and Bayesian methods

2.Martino Bernard, Lightwave circuits for integrated Si photonics

3.Daniel Ricardo dos Santos, Automatic Techniques for the Synthesis and Assisted Deployment of Security Policies in Workflow-based Applications

4.Shahriar Mahbub, Development of innovative tools for multi-objective optimization of energy systems

5.Valentina Marziano, Mathematical modeling for epidemiological inference and public health support

6.Olufemi Akindele Olumodeji, Memristor-based computing architecture with advanced signal processing capabilities

7.Nusrat Sharmin, Correspondence among Connectomes as Combinatorial Optimization

8.Avinash Sudhodanan, Automatic Black-Box Security Testing of Browser-Based Security Protocols

9.Hafeez Ullah, Decoration of graphene sheets with metal and metal oxide nanostructures by low-pressure plasma deposition

[17:30-18:00] Best FBK student award 2016, with talk by the awarded student – Sala Stringa

[18:00-18:30] Alumni reunion – Sala Stringa


Title: My unconventional journey in the AI world

Speaker: Alfio Gliozzo

Abstract: The first time I went to Trento, I was a philosophy undergraduate looking for opportunities to inspire and change my life. After sixteen years, I become a Research Manager in one of the world’s largest AI labs and my research is a subject at the most prestigious universities worldwide. In this talk, I’ll share my unconventional journey in the AI world, illustrating advises and life principles I learnt from my career.

Short bio: Alfio Gliozzo has 16 years of research and development experience in the field of Artificial Intelligence, with strong focus on Natural Language Processing and Semantic Web. He is currently managing the Knowledge Induction department at IBM T.J. Watson Research and teaching Cognitive Computing at Columbia University. He was part of the Deep QA team that build the Jeopardy! system.

Title: European Research Council - Starting Grants: Best ideas for the best researchers

Speaker: Marco Ferraro

Abstract: Through the ERC Starting Grants, the EU attracts young research talent and keeps it in Europe. With the EU backing these grantees will be able to pursue their best ideas, but also create quality jobs for more research staff who wish to work on the frontiers of science. Ultimately they will contribute to creating the most valuable resource Europe has: human capital. The ERC believes in supporting young talent - indeed two thirds of its funding goes to early-career researchers contributing to the future of Europe in terms of science and more broadly. It is pivotal to keep scientific quality as the one and only selection criterion, and to trust researchers to choose significant topics, without imposing any demands on what to be explored. Giving top researchers free reign to follow their scientific curiosity opens the way to real breakthroughs as a recent independent study found: as much as 71% of the first completed ERC projects led to breakthroughs or major scientific advances. This speaks volumes about the relevance to fund bottom-up frontier research – it creates new knowledge and offers new paths for economic growth.

Short bio: Marco Ferraro is graduated in Languages for Interpreting and Translation and has a background experiences in international cooperation. He has a post-graduation master degree on European Project Management and SMEs internationalization processes. Since he joined APRE in September 2013 he has been specializing in international cooperation, Marie Sklodowska-Curie Actions and European Research Council, being the National Contact Point in Italy for both programmes. He is also involved in managing the following projects: Net4Mobility, SiS.net2 (HORIZON 2020) and INCOnet EaP (FP7).

Abstracts of Posters

Title: Security Testing of Android Apps for Detecting Permission Re-delegation Vulnerabilities

Speaker: Biniam Fisseha Demissie

Abstract: Software systems are composed of multiple components. Components hold special privileges to perform different tasks. Controlled access to system resources is achieved through permission based security model. In order to gain privieges, a malicious component usually has to abuse a privileged component. This phenomenon is commonly known as the confused deputy attack. Confused deputy attack occurs when a privileged component performs an activity that needs special permission on behalf of other component that does not have the required permission. Static analysis is often used to detect existence of this vulnerability. However, reports of static analysis are vulnerability points rather than conditions that cause the vulnerability. Therefore, a developer that wishes to fix this vulnerability has to manually analyze the code in order to understand the conditions that cause the vulnerability. Similar to other permission based systems, the Android system also suffers from the confused deputy attack called permission re-delegation. In this work, we present our approach on how to automatically generate test case that reveal permission re-delegation vulnerability in Android apps. Previous results showed that not every reported permission re-delegation is a vulnerability as re-delegation is a feature of the Android framework. Therefore, we need to minimize the report to those potential permission re-delegation vulnerabilities. We propose an automated test oracle based on top applications from the official market and compare behaviors of vulnerable applications to similar applications in the market.

Title: Design for Adaptation of Distributed Service-based Systems

Speaker: Martina De Sanctis

Abstract: Internet of Services applications need to cope with a continuously changing environment, both in terms of the context in which they operate, and of the services, users and providers involved. In this setting, adaptivity is to be considered an intrinsic characteristic of applications rather than an exception to be handled. We propose a design for adaptation approach that fully exploits the advantages of the service- oriented paradigm to support the development and operation of service-based applications operating in highly dynamic environments. The approach is based on dynamic and incremental service composition and re-configuration techniques and it has been evaluated on a real-world scenario in the Smart Cities domain.

Title: Semantic Image Interpretation

Speaker: Ivan Donadello

Abstract: Semantic Image Interpretation (SII) is the process of automatically generating meaningful descriptions of the content of images. An example of meaningful image description is a graph which nodes correspond to image objects and the arcs to semantic relations between objects. Background knowledge (BK), in the form of logical theories, is extremely useful for SII. Many state-of-the-art algorithms for SII mainly adopt a bottom-up approach, which generates semantic interpretations of images starting from their low- level features. BK is used only at a late stage for enriching the semantic descriptions. In this work, we show how BK plays an important role also during the early phase of SII if integrated with low-level image features. To this aim, we propose: (i) a reference framework where a semantic image interpretation is a partial model of the BK. The elements of the partial model are grounded (linked) to a (set of) image segment(s). (ii) The use of Logic Tensor Network to predict an approximation of a partial model of a picture according to the BK. Logic Tensor Network is a framework that integrates learning from numerical data and logical reasoning over the BK. This integration allows us to reason over the low-level features of objects in a picture and to express facts between them consistently with the constraints of the BK. We evaluate our method on the task of classifying objects and their parts in the images of the PASCAL-Part dataset. In this experiment we combine the features extracted with a state-of-the-art object detector with BK expressed as a part-whole ontology. Our approach outperforms the state-of-the-art on object classification, and improves the performances on part-whole relation detection with respect to a rule-based baseline.

Title: Domain Adaptation in Neural Machine Translation

Speaker: Amin Farajian

Abstract: State-of-the-art neural machine translation (NMT) systems are generally very sensitive to the training domain and their performance degrades if the test set belongs to a different domain than the training data. Therefore, the current NMT systems are trained on specific domains by carefully selecting the training sets and applying proper domain adaptation techniques. However, in real-world applications it is very hard, if not impossible, to develop and maintain several specific MT systems for multiple domains. This is mostly due to the fact that usually: i) the target domain is not known in advance, and users might query different sentences from different domains; ii) the application domains are very diverse, making it not feasible to develop and fine-tune one system for each domain, which makes the possibility of developing and fine-tuning one system for each domain unfeasible; iii) there is no (or very limited amount of) indomain training data to train domain-specific MT engines. In this situation, it is necessary to have high quality MT systems that perform consistently well in all (or most of) the domains. In this project we propose effective solutions for developing multi-domain NMT systems that their performance does not degrade by changing the application domain and perform equally well in all the domains.

Title: From lexical opposition to discourse contrast relations: a study for the Italian language with a corpus

linguistics approach Speaker: Anna Feltracco

Abstract: Detecting contrast, both among words and among portions of text, is a fundamental requirement for text analysis, thus playing a crucial role in applications such as machine translation, discourse understanding, and information retrieval. My research is focused on the study of opposition relations between verbs and contrast relations at textual level. Through a crowd-sourcing experiment, I bring evidences that i) opposition relations exist between senses of verbs, and ii) there exist different types of opposition relations with different characteristics. Moreover, I focused on contrast relations at textual level that can be either marked by textual connectives or implicitly conveyed. In particular, I aim at creating an exhaustive list of the textual connectives for contrast relations in Italian, and by analysing corpora I am studying which are other linguistic strategies used to communicate contrast in texts and if these strategies have any regularities (for example, what is the relation between lexical opposition and contrast marked by an explicit connective at textual level?). I mainly investigate these relations in the Italian language, using and developing lexical resources such as the T-PAS resource [1] and the LICO resource [2].


[1]Jezek E., B. Magnini, A. Feltracco, A. Bianchini and O. Popescu, 2014, “T-PAS: A resource of corpus-derived Typed Predicate Argument Structures for linguistic analysis and semantic processing”. In Proceedings of the 9th International Conference on Language Resources and Evaluation (LREC’14), May 26-31, Reykjavik, Iceland, ELRA.

[2]Feltracco A., E. Jezek, B. Magnini and M. Stede, 2016, “LICO: A Lexicon of Italian Connectives.” In

Proceedings of the Third Italian Conference on Computational Linguistics (CLiC-it 2016), December 5-7, Napoli, Italy.

Title: Development of Innovative Materials and Substrates for Gas Sensing Technology

Speaker: Andrea Gaiardo

Abstract: In the last years, the research in the gas sensor field experienced a significant boost. Gas sensors represent the crucial elements in gas monitoring systems and olfactory systems for several applications: environmental monitoring, safety and security, quality control of food production, medical diagnosis and so on. The principal parts that compose a chemoresistive gas sensor, the microheater and the sensing material, are under continuous evolution, in order to improve the whole performance of the device. From the point of view of the gas sensing design, the substrate plays a fundamental role, because acts as a heater, mechanical support and transducer of the sensor response. The application of MEMS technology for the fabrication of silicon device with low power consumption has offered new opportunities for innovative gas sensor design. At the same time, the development of new sensing materials is strongly demanded, in order to exceed the problems showed by the materials proposed until now, such as lack of stability and selectivity.

In this work, we studied different approaches in order to realize suitable silicon microheaters for chemoresistive gas sensors and new sensing materials, available for high operating temperatures (650 °C) through the MEMS technology.

Title: Multifunctional Bio-inspired Surfaces

Speaker: Simone Ghio

Abstract: Mimicking nature to achieve special properties is a brunch of science that has become popular in the last few decades. Of particular interest for their properties are lotus leaf and sharkskin. Lotus leaf has a superhydrophobic surface and sharkskin has drag reduction properties. Specifically the self-cleaning effect is wildly known as the ‘lotus effect’. Bio-inspired surfaces of lotus leaf have low adhesion, water repellency and self-cleaning [1-3]. These properties arise from a combination of chemical and micro-structuration of the surface. Combining roughness and a hydrophobic surface leads to superhydrophobicity, while using a hydrophilic substrate generally leads to superhydrophilicity. Sharkskin is known to have a low drag coefficient that lead at increment of swimming speed and lower power consumption. The second part of my work consist on combining lotus leaf and sharkskin effects. For this purpose a systematic range of hierarchical and non-hierarchical structures have been analysed, in order to find a clear relationship between wettability and flow properties. Static and dynamic contact angle, tilt angle and pressure drop, measured long a channel, are the value used for this analysis. Part of my work consist in generate and analyze silicon-based micro-, nano- and hierarchical patterns [4]. Also PDMS patterns are generate starting from a silicon mold. Wettability analysis are based on the measure of contact angle, tilting angle and advancing & receding angles.


[1].N. Pugno, Mimicking lotus leaves for designing super-hydrophobic/hydrophilic and super- attractive/repulsive hierarchical nanostructured surfaces, Nanomechanics in Italy (2007) 1–9.

[2].E. Lepore, et al., Plasma and thermoforming treatments to tune the bio-inspired wettability of polystyrene, Composites Part B 43 (2012) 681–690.

[3].Zang, Jianfeng, Seunghwa Ryu, Nicola Pugno, Qiming Wang, Qing Tu, Markus J. Buehler, and Xuanhe Zhao. "Multifunctionality and control of the crumpling and unfolding of large-area graphene." Nature materials 12, no. 4 (2013): 321-325.

[4].Ghio, Simone, Giovanni Paternoster, Ruben Bartali, Pierluigi Belluti, Maurizio Boscardin, and Nicola M. Pugno. "Fast and large area fabrication of hierarchical bioinspired superhydrophobic silicon surfaces." Journal of the European Ceramic Society 36, no. 9 (2016): 2363-2369.

Title: Emergency Event Detection and Geographic Propagation Prediction Using Mobile Phone Data

Speaker: Didem Gundogdu

Abstract: Large scale emergency events may have dramatic political, economical and social consequences. These events may result in higher crime rates, spreading of infectious diseases, economic crises, and migration phenomena (e.g. refugees across borders or internally displaced people). Hence, early detection of emergency events, such as riots and natural disasters, is of paramount relevance. Based on the assumption that emergency events influence people communication and mobility patterns, we propose to use Call Detail Records (CDR), collected by mobile telecommunication operators for billing reasons, to detect such events. To this end, first we apply the Markov modulated Poisson process to detect anomalies for hourly intervals. Second, we will examine how the effects of these events propagate among different geographical areas. The preliminary results show that our approach can successfully detected anomalous events (e.g. confrontations between groups) at country scale. By introducing proximity graphs, which are constructed from mobility patterns of users, we expect to be able to reveal the existence of specific patterns for different types of events.

Title: An approach to conifer species classification based on crown structure modeling in high density airborne LiDAR data

Speaker: Aravind Harikumar

Abstract: The knowledge about the species of trees is essential for precision forest management practices. Modern high density airborne Light Detection and Ranging (LiDAR) systems have the ability to acquire large number of LiDAR points, allowing a very detailed characterization of the forest at the individual tree level. In

this context, it is possible to use such LiDAR data for accurate classification of the tree species. In this paper, we consider the specific problem of species classification of trees belonging to the conifer class. However, species classification within a class (e.g., the conifer class) using LiDAR data is a challenging problem when considering the tree external crown characteristics only. The Internal Crown Geometric Features (IGFs) are defined based on a novel internal branch structure model that uses 3D region growing and Principal Component Analysis (PCA) to delineate the branch structure of a conifer tree accurately. Internal crown geometric features are used together with external crown geometric features (EGFs) to with the objective of performing an efficient conifer species classification. Support Vector Machines (SVM) was used for classification and to evaluate the accuracy.

Title: Oracle Problem in Software Testing

Speaker: Gunel Jahangirova

Abstract: The oracle problem remains one of the key challenges in software testing, for which little automated support has been developed so far. The effectiveness of a test case in revealing software faults depends critically on the quality of the oracle. The poster presents a technique for assessing and improving test oracles by reducing the incidence of both false positives and false negatives. We use test case generation to reveal false positives and mutation testing to reveal false negatives. The experimental results show that the fault detection rate of the oracles after improvement increases, on average, by 48.6%.

Title: Real time adaptation of the stimulation protocols for neuroscientific studies

Speaker: Elena Kalinina

Abstract: Neuroscientific experiments investigate the subject’s response to external stimulation. Stimulation categories and the order they appear to the subject are defined in the stimulation protocol. In traditional experiments, when the data is analyzed in the offline mode, protocol is fixed beforehand and does not change in the course of experimental data acquisition. In real-time neuroscientific experiments the data is analyzed in the online mode simultaneously with the acquisition process. Real-time experiments allow to examine a broader range of scientific questions, for instance, how to bring the subject in a certain state or how the outcome of the cognitive task depends on the subject’s state. There is growing interest in experiments of this type, because the community along with the traditional group studies is more and more turining to modelling individuals for the purpose of customized interventions. A key advantage of real-time experimentation is the possibility to adjust the experiment protocol as you go, basing on the subject’s response. This, in turn, requires having adequate models of the subject’s state dynamics and efficient algorithms for the optimal choice of the stimulation/feedback being delivered to the subject. There is a plethora of methods for the protocol adjustment, but still important gaps exist especially when there is need for adaptation in view of a particular experimental goal. We will show how the problem of real-time protocol adaptation can be tackled in the context of a real-time Galvanic Skin Response (GSR) experiment and in a real-time functional magnetic Resonance Imaging (fMRI) experiment.

Title: Automatic Generation and Recommendation of Personalized Challenges in Gamification

Speaker: Reza Khoshkangini

Abstract: Gamification uses game elements and game mechanics in a non-game context in order to reach a certain goal (e.g., incentivizing people to change their behaviors in a specific domain). While gamification is often effective in inducing such behavioural changes, well-known limitations concern retaining the interest of players over the long term, and sustaining the new behaviors promoted through the game. Hence, there is a need to devise techniques to keep users interested and engaged to participate in the gamified system in the long term. We propose an approach based on the Procedural Content Generation of personalized and contextualized playable units that appeal to each player, and make her user experience more varied and individually compelling. To this end, we have built a context-aware recommender system (CARS) to generate and recommend personalized challenges, based on the player’s preference, history, game status and performance. Hereby we describe our approach, and evaluate it using a smart urban mobility game that involved the proposal of weekly challenges to hundreds of citizens/players.

Title: An Intent-based Negotiation Protocol for Application-Aware Network Services

Speaker: Antonio Marsico

Abstract: Internet traffic is generated by a multitude of different applications, everyone characterized by its own connectivity requirements. However, networks usually provision all the traffic in the same shortest paths between a source and a destination, thus ignoring the specific requirements of every application. In order to overcome this issue, the application-centric networking offers a new approach to construct transport networks. The application requirements are taken into account in the whole provisioning process, from the highest level (e.g., IP) to the lowest one (e.g., optical). The communication of connectivity requirements from applications to networks needs to be performed by means of interfaces that abstract the specific commands and the low level details of the network implementation. In this context, the Intent-driven networking aims at simplifying the configuration of programmable networks by shifting the attention from low-level network configurations to high-level goals. By exploiting this paradigm, the applications need only to concentrate on what they want rather than how the network performs their request. We propose an Intent-based protocol for requesting and negotiating a connectivity service with specific requirements between applications and networks. The applications can request a connectivity service and, in case of the amount of network resources cannot satisfy the original request, the network offers several alternative solutions for a reduced service to the application. Moreover, the acceptance or the rejection of one of the proposed alternative solutions is performed automatically by the application.

Title: Haptic Feedback for Learning to Climb

Speaker: Eleonora Mencarini

Abstract: In my PhD research, I explore the expressive abilities of haptic communication, considering the case study of rock climbing. In recent years, research in Human-Computer Interaction has shown increasing interest in the design and development of interfaces that do not rely only on visual displays and vision as sensory modality to interact. In this respect, haptic feedback has gained increasing interest as means to convey information, since it can be perceived by our body through the sense of touch. So far, haptic feedback has been investigated for the acquisition of motor skills (for example in the domains of surgery, instrument playing, and sport) and for emotional communication, trying the reproduce the expressiveness of touch typical of co-located interpersonal communication (such as intimacy, closeness, reassurance, etc.), also in remote communication. Being linked to the sense of touch in order to be perceived and understood, the systems that exploit haptic feedback need to be in direct contact with the body; thus, they are mainly divided in tangible and wearable devices. In my research, I explore the communicative abilities of haptic feedback in the context of rock climbing, an extreme sport that requires both the skillful use of the body and the control over negative emotions. By considering the opportunities and constraints offered by technology, and the users’ needs, values, and motivations, I aim at designing a valuable and usable device for the community of climbers and, at the same time, at investigating the communicative abilities of haptic feedback. As a general framework for my thesis, I adopt a co-evolutionary design approach (Agostini et al. 2000) which divided the design process in three main aspects, namely user research, technology exploration, and concepts generation. According to this framework, the investigation of the three aspects is performed in parallel and alternate moments of divergence, when the investigation goes deep into the details of each of them, to moments of converge, when each aspect informs the others. So far, I have conducted an extensive field study with beginner climbers aimed at identifying a possible design space for haptic communication in climbing. Results have shown that during the learning phase the probability of experiencing negative emotions is higher, due to the lack of skills and of familiarity with the new sensations that the vertical dimension of this sport provides. Currently, instructors address negative emotions by giving suggestions or reassuring the climber. However, often there are contextual problems that hinder the communication between who climbs and who is on the ground, such as distance and noise. These instances suggest a space of design for alternative ways of communication between the actors involved, and it is within this space of design that I explore the opportunities offered by haptic communication in terms of expressivity, usefulness and acceptability. Regarding the design, I conducted two workshops aimed at exploring form factor, body location, purpose, and kind of vibration of a possible haptic wearable device. From the technological perspective, the exploration conducted has led to choose vibration as form of haptic communication and to the development of an early wearable prototype that can be activated through Bluetooth. As final steps of

my research, I plan to conduct a study on the understandability of different vibrations in order to define a vocabulary of instructions useful in the climbing context and to conduct an evaluation of the wearable vibrotactile prototype in ecological setting (i.e. in a climbing course in an indoor gym).

Title: Agreement and Disagreement: Comparison of Points of View in the Political Domain

Speaker: Stefano Menini

Abstract: The automated comparison of points of view between two politicians is a very challenging task, due not only to the lack of annotated resources, but also to the different dimensions participating to the definition of agreement and disagreement. In order to shed light on this complex task, we first carry out a study, on transcriptions of political speeches, to manually annotate the components involved in detecting agreement and disagreement. Then, based on the features and interactions observed in this study, we propose an approach to capture, via supervised classification, divergences and similarities in points of view between different authors or speakers when they deal with the same topic. We do not focus on debates in dialogical form, but we rather consider sets of documents in which politicians may express their position with respect to different topics, both in an implicit or explicit way, like during an electoral campaign.

Title: Social Annotation and User Profiling

Speaker: Yaroslav Nechaev

Abstract: Social Media have become a valuable source of up-to-date knowledge about any fictional and real- life entity including persons, organisations, events and many others. Recently, researchers have started to use this knowledge to solve or improve on a number of tasks across different fields including Natural Language Processing and Image Processing. Many new research directions have emerged as well. In my research, I propose a concept of Social Annotation that is defined as a task of linking named entities in a given text to their corresponding Social Media entities, like, for example, hashtags, user profiles and pages. This task is very similar to Named Entity Linking, where the Social Network is used instead of the Knowledge Base. Social Annotation can, therefore, provide a rich up-to-date context to the input text, making the result much more appealing and complete for further processing and usage. However, dealing with Social Media imposes a number of new challenges including noisiness and scale of data that need to be processed. I aim to overcome such challenges in my research.

Title: Demonstration of a Hybrid SDN/GMPLS Control Plane for Optical Virtual Private Networks with Restoration Capabilities

Speaker: Federico Pederzolli

Abstract: Creating Optical Virtual Private Networks (OVPNs) over existing deployments is becoming an operator concern. This work describes the first hybrid SDN/GMPLS (i.e., centralized/distributed) implementation addressing the problem of virtualizing an optical network and demonstrates, through an emulated testbed, that the system is responsive and is able to provide OVPN-specific restoration mechanisms.

Title: Gas adsorption in graphene-based nanostructures

Speaker: Andrea Pedrielli

Abstract: Nanoporous materials are very interesting for their performance in gas absorption and gas separation. In recent years, various novel materials were proposed to surpass traditional ones such as silicon carbide, polymers, metal-organic framework (MOF). Among these, many are graphene-based materials, which presents high surface-to-volume ratio, good mechanical properties and considerable chemical stability. Graphene-based nanostructures – such as graphene frameworks, graphene-oxide frameworks (GOF), or functionalized graphene foams – can be used for gas adsorption, separation, and purification. The microstructures and chemistry of these novel materials can be tailored to obtain a favourable trade off between gas permeability and selectivity. In this contribution, we present results of Molecular Dynamics and Grand Canonical Monte Carlo computer simulation methods modeling the properties of the most promising of these graphene-based materials for gas absorption separation.

Title: Ontology Learning in the Deep

Speaker: Giulio Petrucci

Abstract: Ontologies are used in computer science to represent knowledge in a formal and unambiguous way, facilitating knowledge reuse and sharing among people and computer systems. Manually encoding the knowledge carried by unstructured sources (e.g., documents, web-pages, and so on) into a formal representation can be costly and time-consuming. Indeed, the knowledge acquisition bottleneck is one of the main obstacles to a wide adoption of semantic web and knowledge management technology. To overcome this problem, several methods and tools have been proposed to automatically extract (that is, learn) ontologies from unstructured sources, thus supporting knowledge engineers in this task. These methods still suffer from severe limitations: either they are limited in the expressiveness of the automatically acquired knowledge or, when targeting the extraction of expressive knowledge, they rely on sets of hand- crafted patterns that are expensive to maintain and evolve. Our goal is to overcome these limitations by designing and training a single Deep Learning based system capable to support the automatic extraction of expressive knowledge in a completely automatic, end-to-end fashion. Results so far have shown that our approach can effectively contribute to the automatic ontology learning problem.

Title: Deep Learning for Distant Speech Recognition

Speaker: Mirco Ravanelli

Abstract: Deep learning technologies have recently revolutionized many research fields , including vision, machine translation, natural language processing, bioinformatics, taking a fundamental leap towards artificial intelligence. This technology is also playing a remarkable role in progressively building machines that understand speech, resulting in several popular applications ranging from web-search to car control and intelligent personal assistants. Unfortunately, most state-of-the-art systems provide a satisfactory performance only in close-talking scenarios, where the user is forced to speak very close to a microphone- equipped device. Considering the growing interest towards speech recognition and the progressive use of this technology in everyday lives, it is easy to predict that in the future users will prefer to relax the constraint of handling or wearing any device to access speech recognition services, requiring technologies able to better cope with distant-talking interactions also in challenging acoustic environments. A prominent limitation of current systems lies in the lack of matching and communication between the various technologies involved in the distant speech recognition process. The main modules are indeed typically trained independently and arranged in a pipeline based on a unidirectional communication between the components. To address such a concern, we propose to replace this pipeline with an architecture based on a novel network of deep neural networks paradigm, where all the modules are jointly trained and better cooperate with each other thanks to a full communication scheme between them. Experiments, conducted using different datasets, tasks and acoustic conditions, revealed that the proposed framework can significantly overtake other competitive state-of-the-art solutions.

Title: AuthZ/N Protocols for Native Apps: Security and Privacy

Speaker: Giada Sciarretta

Abstract: Digital identity (DI) management is a key factor to enable the adoption of innovative digital and physical services. Requirements for a complete DI solution include legal, security and technological aspects. For example, the national laws (e.g., SPID) and European regulations (e.g., eIDAS) prohibit to use social login solutions to access e-health services. The process of defining DI management systems is only at an early stage and solutions capable of addressing all these aspects are missing. In this context, our goal is to propose novel authentication and authorization solutions that satisfy the expected security requirements and protect user privacy, while complying with national and European laws and directives. In particular, we have focused on mobile technologies. Indeed, while for web applications there exist many secure authentication and authorization solutions to protect the user's DI and online resources, solutions for the mobile context are not well consolidated. This is a huge limitation as mobile applications are dominating the market of online services, with a total of 4.9 billion unique mobile subscribers by the end of 2016 (more than 60% of the world population). Our DI solutions are designed and developed in three different contexts: (1) smart city: we have proposed an OAuth-based solution for secure delegated access to online resources. An implementation of this solution has been integrated in the Smart Community Platform of FBK for developing open services in

the Trentino region and is being used daily by up to 13,000 users. (2) e-health: an implementation of our proposed Single Sign-On model is currently tested by the users of TreC (acronym for Cartella Clinica del Cittadino), a personal health record system. TreC is currently deploying a number of Android applications to support self-management and remote monitoring of chronic conditions. The solution we propose will allow patients to access different TreC e-health mobile applications with a SSO experience. (3) single digital market: we are involved in a EIT Digital project called FIDES (acronym for Federated Identity Management System). Its goal is to manage digital identities in a federated way, creating an identity ecosystem that permits the access to digital and physical services in a secure, cross-border and multi-device way.

Title: Modelling Photoinduced Events and Non-Linear Spectroscopy in Complex Multi-Chromophoric System

Speaker: Francesco Segatta

Abstract: Light-Harvesting (LH) Pigment-Protein Complexes in plants and photosynthetic bacteria, constitute the fundamental units through which sunlight is collected. The harvested energy is then redirected towards a Reaction Center by means of extremely efficient Energy Transfer (ET) processes, where its conversion to chemical energy, for the organisms’ self-sustenance, finally occurs. The high complexity of such systems, where multiple interacting chromophores are embedded in a fluctuating protein matrix, makes it extremely difficult to study their response to external excitations, both theoretically and experimentally. In this respect, advances in non-linear electronic spectroscopy with femtosecond time resolution, such as the Two- Dimensional Electronic Spectroscopy (2DES), have provided new insight on energy transfer processes [1], allowing to decongest the overlapping transient spectra. High density of information is obtained, but to accurately disentangle all these features and reach a detailed and reliable map of the ET routes is still a challenge. A solution is to integrate the measurements with theoretical models that become mandatory for experiments interpretation [2].

Here we introduce a fully Quantum-Chemistry based protocol to simulate 2DES spectra of complex multi- chromophoric architectures of known structure. The proposed scheme relies on a QM/MM multi-scale approach [3,4] able to link single chromophoric units with the whole molecular aggregate via a so called Frenkel Exciton Hamiltonian, eventually delivering the system’s manifold of states with an unprecedented accuracy. The level of sophistication obtained, also including molecular vibrations and environmental fluctuations, allows us to simulate and interpret linear and non-linear spectroscopy, providing a deeper insight of the recorded spectral signatures. Application to the LH2 test case, employed here as a challenging playground, shows a remarkable agreement with the experiments and sheds new light on some still debated features of this extensively studied antenna system.


[1]T. Brixner, J. Stenger, H. M. Vaswani, M. Cho, R. E. Blankenship, G. R. Fleming, Nature, 434, 625-628 (205)

[2]I. Rivalta, A. Nenov, G. Cerullo, S. Mukamel, M. Garavelli, Int. J. Quantum Chem., 114, 85-93 (2014)

[3]C. Curutchet, B. Mennucci, Chemical Reviews, acs.chemrev.5b00700 (2016)

[4]S. Jurinovich, L. Viani, C. Curutchet, B. Mennucci, PCCP, 17 (46), 30783-30792 (2015)

Title: Advanced Methods for the Analysis of Very High Resolution Multi-Sensor Optical Images Time-Series

Speaker: Yady Tatiana Solano

Abstract: The use of remote sensing in the analysis and evaluation of environmental changes has become a valuable tool which relevance has increased with the improvement in acquisition sensor technology as well as in the data/image processing algorithms. The capability to understand and detect those changes increases when we consider Very High spatial Resolution (VHR) multispectral images acquired by passive sensors such as IKONOS, QuickBird, GeoEye, WorldView-2 and Pleiades. Nevertheless, the revisit period of each single sensor, their competing orders, and the weather conditions do not always allow the acquisition of proper and continuous information. To mitigate these limitations, it is possible to build time series by considering images acquired by different sensors. The challenge becomes how to deal with this kind of information. Therefore, the aims of this proposal are mainly two: i) multi-sensor multitemporal VHR data homogenization; and ii) change detection for the analysis of VHR multi-sensor multitemporal images.

Keywords: Remote Sensing, Image Processing, Very High Resolution images, Change Detection, Multi-sensor Integration, Time Series Analysis.

Title: “Two days we have passed with the ancients...” From Historical Travel Reports to Information Extraction

Speaker: Rachele Sprugnoli

Abstract: Information Extraction, that is the task aimed at automatically identifying specific pieces of information in unstructured data, is an active area of research in the Natural Language Processing (NLP) community. However resources and automatic systems developed so far have mainly addressed contemporary texts. Finding new approaches to the automatic identification of information also in historical documents can, indeed, assist historians in enhancing their work while dealing with the ever increasing amount of digital sources and, moreover, it can have an impact both on NLP and on Digital Humanities research. In this poster we present a new Information Extraction task applied to historical documents: more specifically we present a new corpus of historical travel guides, a new scheme for the annotation of Content Types (i.e., text passages with specific semantic and functional characteristics) and the results of two preliminary sets of classification experiments. The identification of Content Types may improve the performance of more complex NLP tasks, for example timeline building, by targeting the portions of the documents that are more relevant.

Title: Exploring Sensorial Features for Metaphor Identification

Speaker: Serra Sinem Tekiroglu

Abstract: Language is the main communication device to represent the environment and share a common understanding of the world that we perceive through of our sensory organs. Therefore, each language might contain a great amount of sensorial elements to express the perceptions both in literal and figurative usage. In contrast to being an anomaly, figurative language, especially metaphors within the scope of this study, is pervasive in the common language and should be handled by any NLP application that includes a semantic task. In order to tackle the semantics of figurative language, several conceptual properties are utilized, such as concreteness or imegeability. However, to the best of our knowledge, there is no attempt in the literature to analyze and benefit from the sensorial elements for the figurative language processing. In this research, we propose to exploit sensorial affinity of the words as a feature for metaphor identification task. Additionally, we introduce the automatic detection of synaesthetic metaphors, which is one of the most common metaphoric transfers.

Title: Time-frequency reassignment for acoustic signal processing

Speaker: Georgina Tryfou

Abstract: Acoustic signal processing is the automatic manipulation of acoustic signals, aiming at the recording, storage and generation of acoustic signals, and the extraction and encoding of related information. There is a wide range of target applications, such as audio analysis and synthesis, speech enhancement, audio source separation and signal processing for loudspeakers and microphone arrays. In the core of such systems, the application of the short-time Fourier transform (STFT) results in a representation of the temporal evolution of the frequency components of the acoustic signal, commonly known as the spectrogram. Although the spectrogram is extensively used, it suffers from certain limitations, as for instance the unavoidable trade-off between the time and frequency resolution. In this thesis we study the use of an improved representation, obtained with the method of time-frequency reassignment. According to this method, the traditional spectrogram, as obtained from the STFT, is reassigned to a sharper version called the reassigned spectrogram. In this thesis, we elaborate on approaches that utilize the reassigned spectrogram, and exploit this representation in the context of different applications. Original solutions are presented in two application domains, namely distant speech recognition and melody extraction from polyphonic music signals.

Abstracts of Short Talks

Title: Detecting brain effective connectivity with supervised and Bayesian methods

Speaker: Danilo Benozzo

Abstract: The study of causality has drawn the attention of researches for centuries from many different fields. In particular, nowadays causal inference is a central question in neuroscience and an entire body of research called brain effective connectivity, is devoted to detect causal interactions between distinct brain areas. Brain effective connectivity is typically studied through the analysis of direct measurements of the neural activity, e.g. magneto-electroencephalography (M-EEG) signals, functional MRI (fMRI) signals, single cell recordings etc. The main purpose of my PhD activity is on methods for doing time series causality. More in details, we focus on a well-establish criterion of causality: the Granger criterion, which is based on the concepts of temporal precedence and predictability. The activity that we present is about a new approach to the problem of time series causality that is based on machine learning and, specifically, on learning from examples. Given a set of signals, their causal interactions are estimated by a classifier that was trained on a synthetic dataset generated in order to be representative of the actual context. This approach that we call supervised parametric approach, was implemented by adopting the Granger criterion of causality and compared with the standard parametric measure of causality derived from the same criterion. Moreover, the roles of the feature space and the generative model of the training set are investigated through a simulation study. And an example of real application is shown with rat neural recordings.

Title: Lightwave circuits for integrated Si photonics

Speaker: Martino Bernard

Abstract: Since the developement of modern electronics the dimension of electronic devices, their integration density and their possible applications has changed a lot. The reduction in the base elements dimension and the development of integrated devices allowed the advent of modern computers. Photonics is following a similar route: starting from the development of optical fibers for long-range signal transmission to integrated devices with footprints of a few squared micrometers. The center around which this work revolves is a project financed by the Provincia Autonoma di Trento (PAT) ''On silicon chip quantum optics for quantum computing and secure communications'', SIQURO. Project SIQURO aims at developing an integrated chip that features many different components: a "Lab On a Chip" (LAC) approach, to explore the possibility of non-linear and quantum experiments within a single chip. The idea is to shrink a number of macroscopic tools normally required to perform the experiment, I.E. an optical table with a footprint of m2, within a single silicon chip of few cm2. In particular I've been involved in the realization of an integrated III-V pulsed laser, photonic circuitry for non-linear generation of photon pairs and photonic circuitry for light manipulation. The chosen platform is silicon which allows to adopt (and adapt) the worldwide standard of CMOS technology.

Title: Automatic Techniques for the Synthesis and Assisted Deployment of Security Policies in Workflow- based Applications

Speaker: Daniel Ricardo dos Santos

Abstract: Workflows specify a collection of tasks that must be executed under the responsibility or supervision of human users. Workflow management systems and workflow-driven applications need to enforce security policies in the form of access control and authorization constraints, such as Separation of Duty. Enforcing these policies is crucial to avoid malicious use, but may lead to situations where a workflow instance cannot be completed without the violation of the policy. The Workflow Satisfiability Problem (WSP) asks whether there exists an assignment of users to tasks in a workflow such that every task is executed and the policy is not violated. The WSP is inherently hard, but solutions to this problem have a practical application in reconciling business compliance and business continuity. The main contributions of this work are three: 1. We present a technique to synthesize monitors to solve the run-time version of the WSP. 2. We introduce and solve a new class of problems related to finding execution scenarios that satisfy properties of interest to users. 3. We implement the proposed techniques in two tools: Cerberus for workflow management systems and Aegis for workflow-driven web applications.

Title: Development of innovative tools for multi-objective optimization of energy systems

Speaker: Shahriar Mahbub

Abstract: From industrial revolution to the present day, fossil fuels are the main sources for ensuring energy supply. Negative effects of fossil fuels on environment urge energy planners to integrate renewable energies into the corresponding energy systems. However, large-scale incorporation of renewable energies into the systems is difficult because of intermittent behaviors, limited availability and economic barriers. It requires intricate balancing among different energy producing resources and the syringes among all the major energy sectors. Although it is possible to evaluate a given energy scenario (complete set of parameters describing a system) by using a simulation model, however, identifying optimal energy scenarios with respect to multiple objectives is a very difficult to accomplished. In addition, no generalized optimization framework is available that can handle all major sectors of an energy system. In this regards, we propose a complete generalized framework for identifying scenarios with respect to multiple objectives. The framework is developed by coupling a multi-objective evolutionary algorithm and EnergyPLAN. The results show that the tool has the capability to handle multiple energy sectors together, moreover, a number of optimized trade-off scenarios are identified. The framework opens a door for policy makers to optimize corresponding energy systems in terms of multiple objectives and choose the appropriate one for his/her respective region.

Title: Mathematical modeling for epidemiological inference and public health support

Speaker: Valentina Marziano

Abstract: Mathematical modeling represents a powerful tool with a wide range of applications in epidemiology and public health. They can help disentangling the complexity of transmission of infectious diseases and provide explanations for the trends observed in the data (e.g. recurrent epidemics, seasonality). More importantly, mathematical modeling plays a crucial role in projecting the future trends of infectious disease dynamics and in estimating the impact of possible interventions thus, supporting health authorities in the choice of adequate strategies for the effective prevention and control of infectious cases. In this thesis I investigate different topics of interest for the current computational epidemiology of infectious diseases and public health. First, I study the role of demographic processes in the epidemiology of varicella and Herpes Zoster and I provide a cost-effectiveness evaluation of varicella and Herpes Zoster vaccination in Italy. Second, I use a computational model to detect and quantify the effect of spontaneous behavioral changes on the spatiotemporal dynamics of the 2009 H1N1 influenza pandemic in England. Finally, I investigate the current epidemiology of measles in Italy in order to shed some light on the epidemiological features of vaccine preventable diseases in frameworks characterized by a low circulation of the virus.

Title: Memristor-based computing architecture with advanced signal processing capabilities

Speaker: Olufemi Akindele Olumodeji

Abstract: Memristor-Based Computing Architecture with Advanced Signal Processing Capabilities is aimed at developing advanced processing architectures, exploiting the main characteristics of the memristor, which combines resistive and memory properties. Memristors have been considered as the fourth fundamental passive element joining the resistor, capacitor and the inductor ̶predicted in the early 70s by Chua and first physically demonstrated in 2008 at the Hewlett Packard laboratories. This work sought to exploit the properties of the memristor particularly its application in analogue integrated circuits. Firstly, an electrical model of the memristor was implemented to enable CAD simulation and then several memristor-based analogue ICs applications were investigated. Novel electronic circuits have been investigated, exploiting memristor characteristics, to be the basic building blocks for the realisation of prototypes of a complex memristor-based adaptive network. Some of the novel circuits implemented during this work are - Architectures for interfacing memristive systems with photodetector front-end; Adaptive Background Subtraction for memristor-based image processing algorithms; Memristors and their applications to adaptive vision systems; multi-layer neural network interfaced with memristive devices.

Title: Correspondence among Connectomes as Combinatorial Optimization

Speaker: Nusrat Sharmin

Abstract: Diffusion magnetic resonance imaging (dMRI) data allows the reconstruction of the neural pathways of the white matter of the brain as a set of 3D polylines, by means of tractography algorithms. The

3D polylines are called streamlines and the set of whole streamlines is called tractogram, which represents the structural connectome of the brain. In neurological studies, it is often important to identify the group of streamlines belonging to the same anatomical structure, called tract or bundle, like the cortico-spinal tract or the arcuate fasciculus. The statistical analysis of the diffusion data of tracts is used in multiple applications, for example, to study gender differences , to observe the changes in age and to correlate it with diseases. The extraction of tracts of interest is called the tract segmentation problem. Due to the anatomical variability across subjects, tract segmentation problem are difficult to solve. In this thesis, we propose a novel supervised tract segmentation method, that segments the tract of interest in a new subject using a set of segmented tracts as prior information. Our proposed supervised segmentation approach is based on the concept of the streamline correspondence i.e. to find which streamline in one tractogram correspond to which streamline in the other tractogram. We showed that streamline correspondence can be a powerful principle to transfer the anatomical information of a given bundle from one subject to another one. In the current literature of supervised segmentation, streamline correspondence has been addressed with a nearest neighbour strategy. We observed that finding corresponding streamlines with the nearest neighbour strategy is suboptimal. In the contrary, in this thesis we address the tract segmentation problem as a rectangular linear assignment problem (RLAP), a cornerstone of combinatorial optimization. By investigating the existing optimal solution of the rectangular linear assignment, we adopted the efficient solution for the streamline correspondence problem.

Title: Automatic Black-Box Security Testing of Browser-Based Security Protocols

Speaker: Avinash Sudhodanan

Abstract: This talk is about Blast, a BLAck-box Security Testing tool for discovering security vulnerabilities in the protocols underling security-critical Multi-Party Web Applications (MPWAs). MPWAs rely on core trusted third-party systems–e.g., payment servers, identity providers–and protocols–e.g., Cashier-as-a-Service, Single Sign-On–to deliver business services to users. Blast leverages 6 attack patterns (generalizations of 11 prominent attacks from the literature) to generate attack test cases against the MPWAs under test. Blast has been used to discover 21 previously-unknown security vulnerabilities in prominent MPWAs (e.g. twitter.com, OpenCart v2.1.0.1). The implementation of Blast is based on OWASP ZAP, a widely-used, open-source penetration testing tool. Blast is currently being experimented at SAP and other industrial players are considering its adoption.

Title: Decoration of graphene sheets with metal and metal oxide nanostructures by low-pressure plasma deposition

Speaker: Hafeez Ullah

Abstract: The depositions of various nano-materials onto graphene nanoplatelets (layers of graphene) are ideal for different applications. Deposition by radio frequency (RF) sputtering technique can degrade layers of graphene due to interaction of high energy sputtered particles and plasma species. By optimizing the deposition parameters, the level of damage to graphene layers can be controlled. In this scenario, niobium pentaoxide (Nb2O5) thin films deposited on graphene nanoplatelets (GNPs) powder by varying RF power and process pressures with powder vibration frequency in argon atmosphere. The structural properties of the samples were investigated by X-ray diffraction (XRD), Transmission electron microscopy (TEM) and Raman spectroscopy. For the chemical analysis of the samples X-ray photoelectron spectroscopy (XPS) was used. The interlayer d spacing of the GNPs calculated from the XRD pattern changes with Nb2O5 decoration. By increasing Nb2O5 contents through RF powers and process pressures variation defects on the surface of GNPs increased and basal planes of GNPs gradually loss their initial ordering, decreasing the degree of graphitization. TEM images demonstrated that GNPs decorated with around 5 to10 nm uniform layer of Nb2O5 on their surface were successfully fabricated. The frequencies of the GNPs G and 2D modes were found to undergo red shifts with decoration of Nb2O5. We explained the red shift of GNPs from the Raman frequencies in terms of tensile strain and doping induced by Nb2O5. The XPS data showed that the deposition of Nb2O5 on GNPs, the Fermi level of GNPs shifts downward gradually with increasing Nb2O5 contents due to a p-type doping effect.